A Short Introduction to Grid Computing High Performance Computing Day Texas A&M University May 2, 2007 Steve Johnson Department of Mathematics and Institute for Scientific Computation Texas A&M University Copyright, Steve Johnson, 2007. All rights reserved.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
A Short Introduction toGrid Computing
High Performance Computing DayTexas A&M University
May 2, 2007
Steve JohnsonDepartment of Mathematics
andInstitute for Scientific Computation
Texas A&M University
Copyright, Steve Johnson, 2007. All rights reserved.
What is a Computational Grid?
● Coordinates resources that are not subject to central control
● Uses open standards
● Nontrivial quality of service is achieved
Ian Foster, "What is the Grid? A Three Point Checklist"
Why use a grid?
● Cycle scavenging
● Event driven computing
● Cumulative cluster computing
● General purpose HPC
● Distributed Data Grids
Cycle Scavenging
● Utilize “excess” cycles on an otherwise idle resource.
● Job output is not time critical● SETI@home, Folding@home. Initiated by
the owner of the resource.● Condor (free), United Devices (not free)
– Schedules jobs on idle workstations– Supports checkpoint/restart– Usually for singlenode nonMPI jobs
1GB2GB mem per core, 3TB disk, GigE and IB (35 nodes)
● Immersive Visualization Center– 25' x 8' x 120 degrees rearprojected curved screen– Dell Precision 690, Dual Xeon, 32GB RAM,
2xNVIDIA Quadro 4500
SURAgrid Resources at A&M(coming soon)
● Hydra– IBM p5/575+. 40 nodes, 32GB/node. 21TB
GPFS disk, HPS, 640 total cores, 5TFlops.– 20% of resources for SURAgrid per IBM/SURA
partnership● TexAQS & Windfarm
– AMD and Intel clusters, 84 cores, 2GB/core, GigE, 410GFlops
– support SURA Coastal Ocean Observing and Prediction (SCOOP) program
SURAgrid Applications
● ADCIRC – Coastal Circulation and Storm Surge Model– Part of SURA Coastal
Ocean Observing and Prediction (SCOOP) program
– Managed by RENCI– Data driven ondemand
computing– Uses its own internal
scheduler, portal
Left: ADCIRC max water level for 72 hr forecast starting 29 Aug 2005,driven by the "usual, alwaysavailable” ETA winds.
Right: ADCIRC max water level over ALL of UFL ensemble wind fields for 72 hr forecast starting 29 Aug 2005, driven by “UFL alwaysavailable” ETA winds.
Images credit: Brian O. Blanton, SAIC
ADCIRC info courtesy of Howard Lander, RENCI
SURAgrid Applications
● Wavewatch 3 – event driven coastal modeling– Part of SCOOP project– LSU